ClawHub
Back to skill

Security audit

Src

Security checks for vulnerabilities and agentic risk

Overview

This is a setup guide for routing Codex through local LiteLLM and Vertex AI; it makes persistent manual config changes but shows no hidden code, exfiltration, or deceptive behavior.

Before installing, review the persistent changes carefully: this guide changes Codex's model provider, points traffic at a local LiteLLM proxy, uses Vertex AI credentials, sets an OPENAI_API_KEY placeholder in future shells, and includes a trusted-workspace example. Apply only the paths and trust settings you actually intend to use.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Agent Config Directory Access

High
Category
Agent Snooping
Content
## 2. Codex Configuration

Codex stores its configuration in `~/.codex/config.toml`. You must configure it to point to your local LiteLLM instance and specifically request the `responses` wire API, as Codex has deprecated the `chat` wire API.

Update `~/.codex/config.toml`:
Confidence
90% confidence
Finding
Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Agent Config Directory Access

High
Category
Agent Snooping
Content
## 2. Codex Configuration

Codex stores its configuration in `~/.codex/config.toml`. You must configure it to point to your local LiteLLM instance and specifically request the `responses` wire API, as Codex has deprecated the `chat` wire API.

Update `~/.codex/config.toml`:
Confidence
90% confidence
Finding
Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Agent Config Directory Access

High
Category
Agent Snooping
Content
## 2. Codex Configuration

Codex stores its configuration in `~/.codex/config.toml`. You must configure it to point to your local LiteLLM instance and specifically request the `responses` wire API, as Codex has deprecated the `chat` wire API.

Update `~/.codex/config.toml`:
Confidence
90% confidence
Finding
Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.

Session Persistence

Medium
Category
Rogue Agent
Content
## 1. LiteLLM Configuration

You need to create a `config.yaml` for LiteLLM that drops complex parameters and sets content to simple strings, then routes a `codex` model alias to your Gemini Vertex endpoint.

Create or update your `config.yaml` (e.g., `/app/config.yaml`):
Confidence
60% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Missing User Warnings

Low
Confidence
94% confidence
Finding
The skill instructs users to append an environment variable to their shell profile, which creates a persistent configuration change without clearly warning that it will affect future shells. While the value shown is only a placeholder token, persistent profile edits can have unintended side effects and are riskier than a session-scoped export.

Static analysis

No suspicious patterns detected.